linux/fs/xfs/libxfs/xfs_ag_resv.c

337 lines
9.4 KiB
C
Raw Normal View History

xfs: set up per-AG free space reservations One unfortunate quirk of the reference count and reverse mapping btrees -- they can expand in size when blocks are written to *other* allocation groups if, say, one large extent becomes a lot of tiny extents. Since we don't want to start throwing errors in the middle of CoWing, we need to reserve some blocks to handle future expansion. The transaction block reservation counters aren't sufficient here because we have to have a reserve of blocks in every AG, not just somewhere in the filesystem. Therefore, create two per-AG block reservation pools. One feeds the AGFL so that rmapbt expansion always succeeds, and the other feeds all other metadata so that refcountbt expansion never fails. Use the count of how many reserved blocks we need to have on hand to create a virtual reservation in the AG. Through selective clamping of the maximum length of allocation requests and of the length of the longest free extent, we can make it look like there's less free space in the AG unless the reservation owner is asking for blocks. In other words, play some accounting tricks in-core to make sure that we always have blocks available. On the plus side, there's nothing to clean up if we crash, which is contrast to the strategy that the rough draft used (actually removing extents from the freespace btrees). Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-19 08:30:52 +08:00
/*
* Copyright (C) 2016 Oracle. All Rights Reserved.
*
* Author: Darrick J. Wong <darrick.wong@oracle.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it would be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
*/
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_sb.h"
#include "xfs_mount.h"
#include "xfs_defer.h"
#include "xfs_alloc.h"
#include "xfs_error.h"
#include "xfs_trace.h"
#include "xfs_cksum.h"
#include "xfs_trans.h"
#include "xfs_bit.h"
#include "xfs_bmap.h"
#include "xfs_bmap_btree.h"
#include "xfs_ag_resv.h"
#include "xfs_trans_space.h"
#include "xfs_rmap_btree.h"
#include "xfs_btree.h"
#include "xfs_refcount_btree.h"
xfs: set up per-AG free space reservations One unfortunate quirk of the reference count and reverse mapping btrees -- they can expand in size when blocks are written to *other* allocation groups if, say, one large extent becomes a lot of tiny extents. Since we don't want to start throwing errors in the middle of CoWing, we need to reserve some blocks to handle future expansion. The transaction block reservation counters aren't sufficient here because we have to have a reserve of blocks in every AG, not just somewhere in the filesystem. Therefore, create two per-AG block reservation pools. One feeds the AGFL so that rmapbt expansion always succeeds, and the other feeds all other metadata so that refcountbt expansion never fails. Use the count of how many reserved blocks we need to have on hand to create a virtual reservation in the AG. Through selective clamping of the maximum length of allocation requests and of the length of the longest free extent, we can make it look like there's less free space in the AG unless the reservation owner is asking for blocks. In other words, play some accounting tricks in-core to make sure that we always have blocks available. On the plus side, there's nothing to clean up if we crash, which is contrast to the strategy that the rough draft used (actually removing extents from the freespace btrees). Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-19 08:30:52 +08:00
/*
* Per-AG Block Reservations
*
* For some kinds of allocation group metadata structures, it is advantageous
* to reserve a small number of blocks in each AG so that future expansions of
* that data structure do not encounter ENOSPC because errors during a btree
* split cause the filesystem to go offline.
*
* Prior to the introduction of reflink, this wasn't an issue because the free
* space btrees maintain a reserve of space (the AGFL) to handle any expansion
* that may be necessary; and allocations of other metadata (inodes, BMBT,
* dir/attr) aren't restricted to a single AG. However, with reflink it is
* possible to allocate all the space in an AG, have subsequent reflink/CoW
* activity expand the refcount btree, and discover that there's no space left
* to handle that expansion. Since we can calculate the maximum size of the
* refcount btree, we can reserve space for it and avoid ENOSPC.
*
* Handling per-AG reservations consists of three changes to the allocator's
* behavior: First, because these reservations are always needed, we decrease
* the ag_max_usable counter to reflect the size of the AG after the reserved
* blocks are taken. Second, the reservations must be reflected in the
* fdblocks count to maintain proper accounting. Third, each AG must maintain
* its own reserved block counter so that we can calculate the amount of space
* that must remain free to maintain the reservations. Fourth, the "remaining
* reserved blocks" count must be used when calculating the length of the
* longest free extent in an AG and to clamp maxlen in the per-AG allocation
* functions. In other words, we maintain a virtual allocation via in-core
* accounting tricks so that we don't have to clean up after a crash. :)
*
* Reserved blocks can be managed by passing one of the enum xfs_ag_resv_type
* values via struct xfs_alloc_arg or directly to the xfs_free_extent
* function. It might seem a little funny to maintain a reservoir of blocks
* to feed another reservoir, but the AGFL only holds enough blocks to get
* through the next transaction. The per-AG reservation is to ensure (we
* hope) that each AG never runs out of blocks. Each data structure wanting
* to use the reservation system should update ask/used in xfs_ag_resv_init.
*/
/*
* Are we critically low on blocks? For now we'll define that as the number
* of blocks we can get our hands on being less than 10% of what we reserved
* or less than some arbitrary number (maximum btree height).
*/
bool
xfs_ag_resv_critical(
struct xfs_perag *pag,
enum xfs_ag_resv_type type)
{
xfs_extlen_t avail;
xfs_extlen_t orig;
switch (type) {
case XFS_AG_RESV_METADATA:
avail = pag->pagf_freeblks - pag->pag_agfl_resv.ar_reserved;
orig = pag->pag_meta_resv.ar_asked;
break;
case XFS_AG_RESV_AGFL:
avail = pag->pagf_freeblks + pag->pagf_flcount -
pag->pag_meta_resv.ar_reserved;
orig = pag->pag_agfl_resv.ar_asked;
break;
default:
ASSERT(0);
return false;
}
trace_xfs_ag_resv_critical(pag, type, avail);
/* Critically low if less than 10% or max btree height remains. */
return avail < orig / 10 || avail < XFS_BTREE_MAXLEVELS;
}
/*
* How many blocks are reserved but not used, and therefore must not be
* allocated away?
*/
xfs_extlen_t
xfs_ag_resv_needed(
struct xfs_perag *pag,
enum xfs_ag_resv_type type)
{
xfs_extlen_t len;
len = pag->pag_meta_resv.ar_reserved + pag->pag_agfl_resv.ar_reserved;
switch (type) {
case XFS_AG_RESV_METADATA:
case XFS_AG_RESV_AGFL:
len -= xfs_perag_resv(pag, type)->ar_reserved;
break;
case XFS_AG_RESV_NONE:
/* empty */
break;
default:
ASSERT(0);
}
trace_xfs_ag_resv_needed(pag, type, len);
return len;
}
/* Clean out a reservation */
static int
__xfs_ag_resv_free(
struct xfs_perag *pag,
enum xfs_ag_resv_type type)
{
struct xfs_ag_resv *resv;
xfs_extlen_t oldresv;
int error;
trace_xfs_ag_resv_free(pag, type, 0);
resv = xfs_perag_resv(pag, type);
pag->pag_mount->m_ag_max_usable += resv->ar_asked;
/*
* AGFL blocks are always considered "free", so whatever
* was reserved at mount time must be given back at umount.
*/
if (type == XFS_AG_RESV_AGFL)
oldresv = resv->ar_orig_reserved;
else
oldresv = resv->ar_reserved;
error = xfs_mod_fdblocks(pag->pag_mount, oldresv, true);
resv->ar_reserved = 0;
resv->ar_asked = 0;
if (error)
trace_xfs_ag_resv_free_error(pag->pag_mount, pag->pag_agno,
error, _RET_IP_);
return error;
}
/* Free a per-AG reservation. */
int
xfs_ag_resv_free(
struct xfs_perag *pag)
{
int error;
int err2;
error = __xfs_ag_resv_free(pag, XFS_AG_RESV_AGFL);
err2 = __xfs_ag_resv_free(pag, XFS_AG_RESV_METADATA);
if (err2 && !error)
error = err2;
return error;
}
static int
__xfs_ag_resv_init(
struct xfs_perag *pag,
enum xfs_ag_resv_type type,
xfs_extlen_t ask,
xfs_extlen_t used)
{
struct xfs_mount *mp = pag->pag_mount;
struct xfs_ag_resv *resv;
int error;
resv = xfs_perag_resv(pag, type);
if (used > ask)
ask = used;
resv->ar_asked = ask;
resv->ar_reserved = resv->ar_orig_reserved = ask - used;
mp->m_ag_max_usable -= ask;
trace_xfs_ag_resv_init(pag, type, ask);
error = xfs_mod_fdblocks(mp, -(int64_t)resv->ar_reserved, true);
if (error)
trace_xfs_ag_resv_init_error(pag->pag_mount, pag->pag_agno,
error, _RET_IP_);
return error;
}
/* Create a per-AG block reservation. */
int
xfs_ag_resv_init(
struct xfs_perag *pag)
{
xfs_extlen_t ask;
xfs_extlen_t used;
int error = 0;
/* Create the metadata reservation. */
if (pag->pag_meta_resv.ar_asked == 0) {
ask = used = 0;
error = xfs_refcountbt_calc_reserves(pag->pag_mount,
pag->pag_agno, &ask, &used);
if (error)
goto out;
xfs: set up per-AG free space reservations One unfortunate quirk of the reference count and reverse mapping btrees -- they can expand in size when blocks are written to *other* allocation groups if, say, one large extent becomes a lot of tiny extents. Since we don't want to start throwing errors in the middle of CoWing, we need to reserve some blocks to handle future expansion. The transaction block reservation counters aren't sufficient here because we have to have a reserve of blocks in every AG, not just somewhere in the filesystem. Therefore, create two per-AG block reservation pools. One feeds the AGFL so that rmapbt expansion always succeeds, and the other feeds all other metadata so that refcountbt expansion never fails. Use the count of how many reserved blocks we need to have on hand to create a virtual reservation in the AG. Through selective clamping of the maximum length of allocation requests and of the length of the longest free extent, we can make it look like there's less free space in the AG unless the reservation owner is asking for blocks. In other words, play some accounting tricks in-core to make sure that we always have blocks available. On the plus side, there's nothing to clean up if we crash, which is contrast to the strategy that the rough draft used (actually removing extents from the freespace btrees). Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-19 08:30:52 +08:00
error = __xfs_ag_resv_init(pag, XFS_AG_RESV_METADATA,
ask, used);
if (error)
goto out;
}
/* Create the AGFL metadata reservation */
if (pag->pag_agfl_resv.ar_asked == 0) {
ask = used = 0;
error = xfs_rmapbt_calc_reserves(pag->pag_mount, pag->pag_agno,
&ask, &used);
if (error)
goto out;
xfs: set up per-AG free space reservations One unfortunate quirk of the reference count and reverse mapping btrees -- they can expand in size when blocks are written to *other* allocation groups if, say, one large extent becomes a lot of tiny extents. Since we don't want to start throwing errors in the middle of CoWing, we need to reserve some blocks to handle future expansion. The transaction block reservation counters aren't sufficient here because we have to have a reserve of blocks in every AG, not just somewhere in the filesystem. Therefore, create two per-AG block reservation pools. One feeds the AGFL so that rmapbt expansion always succeeds, and the other feeds all other metadata so that refcountbt expansion never fails. Use the count of how many reserved blocks we need to have on hand to create a virtual reservation in the AG. Through selective clamping of the maximum length of allocation requests and of the length of the longest free extent, we can make it look like there's less free space in the AG unless the reservation owner is asking for blocks. In other words, play some accounting tricks in-core to make sure that we always have blocks available. On the plus side, there's nothing to clean up if we crash, which is contrast to the strategy that the rough draft used (actually removing extents from the freespace btrees). Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-09-19 08:30:52 +08:00
error = __xfs_ag_resv_init(pag, XFS_AG_RESV_AGFL, ask, used);
if (error)
goto out;
}
out:
return error;
}
/* Allocate a block from the reservation. */
void
xfs_ag_resv_alloc_extent(
struct xfs_perag *pag,
enum xfs_ag_resv_type type,
struct xfs_alloc_arg *args)
{
struct xfs_ag_resv *resv;
xfs_extlen_t len;
uint field;
trace_xfs_ag_resv_alloc_extent(pag, type, args->len);
switch (type) {
case XFS_AG_RESV_METADATA:
case XFS_AG_RESV_AGFL:
resv = xfs_perag_resv(pag, type);
break;
default:
ASSERT(0);
/* fall through */
case XFS_AG_RESV_NONE:
field = args->wasdel ? XFS_TRANS_SB_RES_FDBLOCKS :
XFS_TRANS_SB_FDBLOCKS;
xfs_trans_mod_sb(args->tp, field, -(int64_t)args->len);
return;
}
len = min_t(xfs_extlen_t, args->len, resv->ar_reserved);
resv->ar_reserved -= len;
if (type == XFS_AG_RESV_AGFL)
return;
/* Allocations of reserved blocks only need on-disk sb updates... */
xfs_trans_mod_sb(args->tp, XFS_TRANS_SB_RES_FDBLOCKS, -(int64_t)len);
/* ...but non-reserved blocks need in-core and on-disk updates. */
if (args->len > len)
xfs_trans_mod_sb(args->tp, XFS_TRANS_SB_FDBLOCKS,
-((int64_t)args->len - len));
}
/* Free a block to the reservation. */
void
xfs_ag_resv_free_extent(
struct xfs_perag *pag,
enum xfs_ag_resv_type type,
struct xfs_trans *tp,
xfs_extlen_t len)
{
xfs_extlen_t leftover;
struct xfs_ag_resv *resv;
trace_xfs_ag_resv_free_extent(pag, type, len);
switch (type) {
case XFS_AG_RESV_METADATA:
case XFS_AG_RESV_AGFL:
resv = xfs_perag_resv(pag, type);
break;
default:
ASSERT(0);
/* fall through */
case XFS_AG_RESV_NONE:
xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, (int64_t)len);
return;
}
leftover = min_t(xfs_extlen_t, len, resv->ar_asked - resv->ar_reserved);
resv->ar_reserved += leftover;
if (type == XFS_AG_RESV_AGFL)
return;
/* Freeing into the reserved pool only requires on-disk update... */
xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FDBLOCKS, len);
/* ...but freeing beyond that requires in-core and on-disk update. */
if (len > leftover)
xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, len - leftover);
}